The End of Junior Engineering

Anthropic doesn't hire junior engineers anymore. The co-founder said so on a podcast last week, in plain English, and the only surprising thing is how few people heard it.

Here is the quote, from Jack Clark on The Reason Interview With Nick Gillespie on June 24:

We're hiring more people with lots and lots of experience than we did before, because the returns on intuition are much greater than before, because now you don't do the schlep work to run your experiments. ... Previously we needed to also give you an engineering team so that you and the engineers could run the experiments. Now Claude runs the experiments, so actually let's hire way more people with, like, senior intuition than we did before, because we don't need to scale these or engineers around them.

Read it twice. The interesting word is not "Claude." It's "intuition."

Anthropic is not telling us that AI is faster than a junior engineer. Everyone already knows that. Anthropic is telling us that the entire economic logic of bundling one senior researcher with a team of more junior people has collapsed. The senior's ideas used to be the constraint; the bottleneck was throughput. Now Claude is the throughput. So the senior researcher hires more senior researchers, and the junior engineer slot just stops being staffed.

That is not "AI takes the jobs." It is a particular kind of restructuring, and it has a specific failure mode that most of the takes are missing.

The data is starting to back the anecdote

Clark's statement is one executive on one podcast. On its own it would not be enough. The Stanford Digital Economy Lab paper from Bharat Chandar, Erik Brynjolfsson, and Ruyu Chen is the part that turns it into a pattern.

They used ADP payroll data, which is the actual paychecks of millions of workers, to look at hiring by age and occupation since LLMs went mainstream. The headline finding: workers aged 22 to 25 in the most AI-exposed jobs have seen a 16 percent relative decline in employment. Older workers in the same fields are stable or growing. Workers in less-exposed jobs are fine across the board.

The within-firm number is 13 percent. That is what you get when you control for whatever else was happening at each company. Same firm, same year, same hiring climate: junior hiring in AI-exposed roles fell 13 percent against junior hiring in non-exposed roles inside the same payroll. Replications using Revelio Labs workforce data in the US and administrative records in the UK show similar patterns. Danish data does not, and honest researchers will tell you the international picture is messy.

But the direction is clear, the mechanism is clear, and now a frontier AI lab has volunteered that yes, this is exactly what they are doing on purpose.

The problem is the ladder, not the rung

This is where most takes fall apart. The headline becomes "AI replaces junior devs," and then everyone fights about whether AI can really replace a junior dev, whether the new grads were any good in the first place, and whether bootcamps are dying.

That fight is mostly noise. The actual problem is structural.

Junior engineers become senior engineers. That is what they're for.

The apprenticeship is the point. The reason a company hires juniors is not primarily that they're cheap labor for boilerplate. It's that some fraction of them will, in five to seven years, have walked enough on-call rotations and shipped enough broken migrations to become the senior engineer who runs the team. The cheap labor is a side effect. The pipeline is the asset.

When Anthropic says it now hires "senior intuition" instead of an engineering team, what it has actually done is delete a rung from the ladder. That rung is where senior intuition comes from in the first place. You can run that org structure as long as the rest of the world is still training people you can poach later. You cannot run that org structure on a planet where everybody has done the same thing.

If the Stanford trend holds and Anthropic's posture spreads, the next senior cohort atrophies by attrition. Not all at once. Just quietly, year after year, the pipeline narrows. In 2030 the senior engineers are 30 percent more expensive than they should be and every company is trying to poach the same two hundred people.

What founders should actually do with this

Three things, none of them "freeze junior hiring."

First, do not blindly copy Anthropic. Anthropic is a research lab whose org is built around frontier AI research and whose product is the very thing absorbing the entry-level work. Their hiring posture optimizes for a problem most companies do not have. Treating their org chart as a template is the same mistake people made copying Google's interview process in 2012, when Google was running a funnel for a one-percent acceptance rate and the rest of us were running a funnel for everyone else.

Second, rebuild the apprenticeship around AI rather than removing it. The junior engineer who now writes most of their code with Claude or Cursor is not the same job as the 2018 junior engineer, and pretending otherwise is how you end up with juniors who can ship features and cannot debug their own logs. Give juniors actual stakes: on-call rotations, real incident reviews, customer calls, postmortems they have to defend out loud. The thing AI can't do for them is the part that makes them senior.

Third, if AI is genuinely the new junior on your team, treat it like one. That means review, accountability, audit trails, and a human who owns the outcome. At Osano we think about this for the privacy and governance side specifically. When the model becomes the contributor that touches the database, the access log, and the user record, the question "who pressed enter" stops having an easy answer. The companies that let the model push to prod with no human in the loop are running an experiment they have not modeled. Some of them will find out the hard way that "the agent did it" is not a defensible answer in a regulatory letter.

The window is short and the signal is loud

Clark himself said the part that should make every founder uncomfortable: AI might deliver above-trend GDP growth at the same time as recession-level unemployment, and no government has planned for that. He isn't predicting it with certainty. He is saying scenario-plan for it.

So should you.

The next two hiring cycles will calcify whatever pattern you set. Hire the way Anthropic hires and you'll have a senior-heavy team that scales beautifully on routine work and falls over the first time the org needs someone new with operator instincts. Cut juniors entirely and you'll buy your senior bench from your competitors at a 50 percent premium in 2028. Keep hiring juniors but give them the same scoped 2018 work, and you'll be paying for output you could get from a model.

The interesting move is the third one nobody is selling: hire juniors, give them serious responsibility on day one, pair them with the model rather than around it, and explicitly design the next-rung work that builds judgment.

Anthropic just told us, with admirable directness, that it does not need junior engineers. Most of the founders I talk to are going to read that as permission. The smarter read is that they just told you exactly how the senior-engineer market will look in 2030.

Please tell me why I'm wrong.